Physical AI Engineering Lead

Job Description

Why this role exists

Most AI still lives behind a screen. We're building the layer where it picks things up, walks across a factory floor, and gets things wrong in ways a chatbot never can.

WIN runs a working Physical AI lab. You'll have day-to-day access to a multi-site fleet spanning industrial cobots, mobile robots and AMRs, quadrupeds, and humanoids — including Universal Robots and Franka arms alongside Unitree platforms — plus a bench of end-effectors and low-cost research rigs for fast iteration.

You'll also inherit real client problems in manufacturing, logistics, and facilities that need someone who can turn frontier robot learning research into something that survives contact with a customer site.

We're looking for the lead engineer who wants to own that whole path: model, simulator, robot, deployment, and the team that builds it.

͏

What you'll own

The technical architecture. Define the reference stack for Physical AI at Wipro — perception, policy, planning, control, and the data layer underneath it. Make the calls on what we build, what we buy, and what we adopt from the open ecosystem.

Sim-to-real, end to end. Build the simulation and digital twin pipelines that let us train and validate policies at scale, then close the gap on real hardware. Domain randomization, synthetic data generation, evaluation harnesses that tell the truth.

Robot learning on real fleets. Lead work on vision-language-action models, imitation learning from teleoperation data, and world models — applied to arms, mobile bases, and humanoids in our lab and at client sites.

Client-facing engineering. Scope and lead proof-of-value engagements, translate a customer's messy operational reality into a tractable robotics problem, and stay close enough to the build that your estimates are credible.

A multi-disciplinary team. Hire, mentor, and grow engineers and research scientists across ML, robotics, controls, and simulation. Set the engineering bar and defend it.

Ecosystem depth. Work directly with robot OEMs, foundation-model providers, chip vendors, and hyperscalers to shape joint roadmaps and keep our stack ahead of the commodity curve.

͏

What we're looking for

• Strong hands-on foundation in AI/ML and robotics, including modern robot learning approaches — imitation learning, policy learning, or VLA architectures. You can read a policy-learning paper and also debug a TF tree at 11pm.
• Solid grounding in kinematics, motion planning, and control (MoveIt, OMPL, or equivalent), with Python and C++ fluency in a real-time context — you've written code that runs inside a control loop, not just around one.
• Production or research depth in ROS 2, and simulation environments such as Isaac Sim / Isaac Lab, MuJoCo, Gazebo, or Unity.
• Experience building teleoperation and data-collection pipelines — trajectory capture, dataset curation, and the tooling that makes a demonstration corpus usable for training.
• Demonstrated experience taking a robotics or embodied AI system from prototype to something that ran reliably outside the lab.
• Track record of technical leadership: architecture decisions, code review culture, and engineers who got better because they worked for you.
• Comfort in front of customers and executives. You can explain a sim-to-real gap to a plant head and a business case to a CXO in the same afternoon.

͏

Bonus points

• Master's or Ph.D. in Robotics, AI, ML, or a related field.
• Published work in robot learning, manipulation, or locomotion.
• Familiarity with VLA models (OpenVLA, π0, Gemini Robotics class) or LeRobot.
• Deployment under tight hardware constraints: on-device inference, latency budgets, Jetson-class compute.
• Manipulation-heavy experience: grasping, force control, dexterous end-effectors.
• Background in consulting, product innovation, or enterprise AI delivery — you've shipped under someone else's constraints.
  
  

  What you get

  A lab with real hardware and a funded roadmap behind it. Deep working relationships across the robot and model ecosystem. And a mandate to build something rather than benchmark it forever.